Using Space Effectively: 2D

From CS294-10 Visualization Sp11

Jump to: navigation, search

Lecture on March 7, 2011

Slides

Contents

Readings

  • Multi-Scale Banking to 45 Degrees. Heer & Agrawala. (pdf)
  • Pad++: A zooming graphical interface for exploring alternate interface physics, Bederson & Hollan (acm)
  • Chapter 11: The Cartogram: Value-by-Area Mapping. In Cartography: Thematic Map Design. Dent (pdf)

Optional Readings

  • Generalized fisheye views, Furnas. (acm)
  • Hyperdimensional data analysis using parallel coordinates, Wegman (jstor)
  • A framework for unifying presentation space, Carpendale & Montagnese. (acm)
  • Nomography
  • Cartogram central

Julian Limon - Mar 07, 2011 08:31:17 pm

I was intrigued by the whole concept of banking to 45 degrees in today's lecture. I have to admit I didn't read the paper beforehand, so I found it quite surprising. I really like how Heer and Agrawala challenged Cleveland's techniques and proposed new and easier baking techniques to determine the most effective ratio in a visualization. Although I was lost in some of the mathematical details, I found their argument really compelling.

This idea of ratios made me think of Tufte. In one of his books, Tufte mentioned that only a small portion of Playfair's diagrams are taller than wider. He didn't go into the details of analyzing angles or measuring human perception, but apparently Playfair did a great job in intuitively determining the right ratio to convey his stories.

And speaking of stories, I appreciated how clear Maneesh was today when he argued that every visualization makes a statement. We have been told otherwise and we tend to easily criticize visualizations that omit the zero point or drive the attention to certain portions of the data. It is refreshing to hear someone who still believes that visualizations are an opinionated expression (abstraction?) about a more complex story.

Michael Hsueh - Mar 08, 2011 04:30:54 am

I liked the Dent piece on cartograms. I didn't know the actual term for value-by-area maps until I read it. I have never been a huge fan of these types of maps. The fact that a map is used at all suggests that geospatial reasoning is at least relevant to the data being shown (otherwise why use a geographic map?). That being true, altering core geographic features (shape/size/position) can undermine some of the information and reasons for using a geographic map in the first place. Still, I can see how cartograms are very useful in some cases (for example, area can probably effectively encode more distinct values than a simple color gradient).

I thought an important disadvantage of cartograms is the difficulty of their construction. Dent outlines a construction method which involve piecing together finely sized units into shapes and arranging the shapes to roughly fit together. He acknowledges that experimentation is central to the process, which is also likely time consuming. If no appropriate arrangement is found, then non-contiguous maps can be used. Dent mentions the possibility of using computers, but is weary of the tendency for automation result in quantity over quality. I would say that computers are rather necessary now for creating cartograms that are scalable and accurate. Given the experimental nature of cartogram design, the speed afforded by computers is perhaps necessary for this method to be viable from a time perspective.

Dan - Mar 08, 2011 10:32:05 am

Cartography paper: The thematic paper presented different techniques for visualizing data associated with geographical space. I thought it was interesting when they skewed the sizes of states to show different area-value representations. This was a bit strange for me, especially when New Jersey and Delaware seemed huge! It was hard to interpret wheth er the entire size was what to look at, or the size compared to its normal size.

The pad++ paper presented a new system for navigating and exploring different types of data, implemented the zooming metaphor. A paper which in my opinion focused on many technical aspects of running time instead of actual interface concepts, but nevertheless a valuable read. The efficiency methods they implemented seemed to be more related to computer graphics than visualization or information transmission: spatial indexing, restructuring, spatial level-of-detail, clipping, and refinement. However, this is key to navigating and zooming through large datasets! And given the time, 1994, perhaps this was what a great focus.

Multi-scale banking to 45 degrees: A great idea, using aspect ratio to skew analysts' perceptions of visualized data. Clearly a million dollar idea that should be used in accounting of large financial firms. Cleveland wants us not to worry necessarily about the zero, and fill the space of a graph using the aspect ratio. I do think its a good idea to utilize the screen real-estate, or in this case, graph real-estate to get more out of data. We can also maximize discriminablility by keeping the average of line segments to 45 degrees. There are other techniques as well, which were discussed in great mathematical detail.

Michael Cohen - Mar 08, 2011 02:39:26 pm

Regarding cartograms: one disadvantage that occurred to me (and that I was surprised not to see mentioned in the reading) is that readers familiar with the geography being depicted may interpret the magnitude of the data as being proportional to the degree of exaggeration, rather than the actual size of the area on the map. For example, being from Massachusetts, when I looked at the traffic congestion map I found myself thinking, "wow, Massachusetts has a lot of congestion, look how blown up it is!" Then I looked at California and thought, "that looks more like its 'normal' size, so its congestion must be moderate, on average." Then I caught myself and realized that both states were being depicted at about the same size, meaning they actually have about the same amount of congestion.

So I would argue that by using familiar shapes rather than generic bars, cartograms introduce another perceptual layer (our memory of the "standard" shape) that may distort the point being communicated.

Krishna - Mar 08, 2011 03:53:36 pm

It wasn't completely clear to me as to what the take aways are from the Pad++ paper. As Dan pointed out, the focus of the paper tended towards implementation strategies rather than their visualization idea. Also, I am not sure on how their zooming interface would visually scale with large amounts of information - as the information scales up, each data point would tend to occupy smaller and smaller spaces and it could get really difficult to precisely zoom a particular area on the screen. Maybe the strategy would work when coupled with a filtering mechanism. I like their idea of incorporating wear and tear for digital artifacts. Although, wear and tear information can be used to show the frequently used (or) handled artifacts, it may not project the relevance of an artifact for a given query. In other words, wear and tear are interesting features of data but I am not sure how they can be used to build exploratory visualization tools in a generic sense.

On cartograms, I think non-contiguous cartograms are better even though information on contiguity is lost. Because, the Map is distorted anyway I don't think this would be a serious disadvantage. Moreover, by only approximating the postion of the regions with respect to other regions any transformation of the region's shape or size does not seriously affect the perception of the user unlike in the case when the regions are connected - at least this is how I felt while trying to understand the examples given in that chapter.

Brandon Liu - Mar 08, 2011 04:29:08 pm

I was curious about cartograms outside of the United States. The paper specifically mentions that cartogram literacy is tied to the reader's familiarity with the data represented: for this reason, the 50 states seem to be the exception, rather than the rule, for how unique geographic shapes are. Here's some other cartograms: World Population http://www.nimblebooks.com/wordpress/2006/08/rectangular-cartogram-of-world-population/ World Oil http://www.physics.unc.edu/~cecil/travels/Energy%20stuff/CARTOGRAMS/WorldOilCartogram1.png Southern Michigan http://indiemaps.com/images/williamBunge/detroit_cartogram.png Europe http://www.webmapper.net/img/blog/eucartogram2009.png I had a lot more trouble recognizing countries in the EU cartogram; this is mostly because I'm less familiar with the geography, but also because the shapes are less distinctive.

Matthew Can - Mar 08, 2011 03:41:50 pm

The Pad++ graphical interface provides an interesting way of navigating large information spaces. All of the information is present on one display, though rendered at different scales as the user zooms in and out. My sense is that this technique works well for structured, hierarchical information. For example, by progressive zooming, a user can view population statistics at the national, state, and county levels. However, I'm skeptical that this is a good technique for visualizing data that doesn't follow a hierarchy. With this technique, it's also difficult to compare two regions of the data at fine scales (the hypertext support might allow this, but that requires building the links first).

More generally, the authors mention that Pad++ fits into their larger agenda of an informational physics approach to interface design. It's a neat concept to rearrange or modify digital objects according to user interaction.

Jessica Voytek - Mar 08, 2011 06:03:17 pm

The discussion of nomograms in class yesterday really caught my attention. They are an elegant solution to what would otherwise require time consuming mathematical calculation. My question is whether they have much relevance today with the prevalence of mobile and embedded computers. I've been trying to think of the criteria that might point to the use of nomograms rather than embedded computers. The first that comes to mind obviously is an absence or small number of computers, as in a natural or man-made disaster relief. In this case the ability to print and distribute many cheap calculating nomograms might be useful or even necessary. The second criteria includes situations in which an embedded computer able to withstand the pressures of the environment make it more expensive or less reliable than one would hope, say for scuba diving (although there are a number of commercial dive computers available). Any more?

Saung Li - Mar 08, 2011 07:04:59 pm

I was also wondering about whether nonograms are used today. The simplicity of the nonogram used by sailors to calculate speed, distance, or time fascinated me, as this is something they had to rely on without calculators. Now that calculators and mobile phones are so widespread, nonograms might have lost their importance. With regards to multi-scale banking, I find that this is a good way to tackle aspect-ratios. I still find it a bit difficult to see which aspect-ratio is good, as that depends on what we are trying to look for. Different ratios could lead to different conclusions that users make when looking at the visualization, and these could end up misleading people. I like the idea of sparklines a lot, as they are such small line graphs that can reveal a lot of general information about something such as stocks. This can be used as small multiples or embedded within bodies of text to make details more compact. This is another example showing how people can view such small details in graphics.

Sally Ahn - Mar 09, 2011 01:10:38 am

Michael Cohen brought up a good point about cartograms that the relative sizes of the actual areas of the chosen geographical context may confuse or bias the user in interpreting the visualized data. Overall, I think cartograms are interesting visualizations, but more vulnerable to misinterpretation than most. Dent clearly notes that cartograms can be misinterpreted. I think the best usage of cartograms would be in communicating unexpected data that leverages the viewer's geographical knowledge to "shock" him with "spatial peculiarities," as Dent describes.

The Pad++ interface allows users to view large data at multiple scales by zooming. I think using data semantics to determine the best data to reveal at each scale level and users' interaction (as in the html viewing example) as parameters for the final visualization is a good idea. The informational physics approach brings up questions on the types of metaphors that exist in interface design. I didn't quite follow their argument against metaphors that become "ubiquitous" and "die." The authors state that "successful metaphors also wind up as dead metaphors," and names the ubiquitous files, menus, and windows interface as an example. Why should metaphors lose value as their popularity increases; shouldn't the familiarity of the metaphor only focus the users' attention more on the data? From the designers' point of view, it is important to consider the data itself rather than trying to fit the data into existing and familiar interfaces, but all advantages of the chosen metaphor are lost if the user cannot recognize it.

Thomas Schluchter - Mar 08, 2011 11:23:18 pm

I was intrigued by how changing aspect ratio in some visualizations (mostly line charts, it appears) can really influence what we perceive. The example of the vertically condensed sun radiation chart in the slides revealed a morphology of curves that wasn't very visible in the stretched version.

This relates back to the point that every visualization tells a story. The whole discussion about lying with statistics mostly centers around contentious issues that are manipulatively represented to make a (perhaps ideologically motivated) point. This creates the false impression that there is an unbiased, truthful way of representing data, and suggests that playing with scale, aspect ratio, labels etc. always detracts from this truthful way. I think that visualizations for scientific purposes should be allowed greater freedoms in this regard as long as they conform to the standards of scientific reasoning.

Systematically manipulating a graph's appearance can help emphasize relationships in data that would otherwise be difficult to make explicit. As long as the integrity of the data represented is maintained, these techniques of manipulation contribute to the generation of insights rather than obscure the "facts".

Siamak Faridani 02:01, 9 March 2011 (CST)

Reading the paper on banking to 45 degrees made me think about what does it really mean to have a fair and unbiased visualization? In terms of perception, to me the only way to present data without any bias is just providing the raw data. Anything else is biased. Another thing was the sparklines, from Tufte's point of view the aspect ratio of a sparkline is pretty much fixed (it needs to fit within two lines of a newspaper) it makes me think that a sparkline is already biased towards a specific type of data representation.

The second challenge that I have been thinking about lately (and after struggling with protovis) is? what are the flow control statements that a visualization language should have? what I am saying is that in protovis I cannot generate panel visualizations with a for loop! I cannot do a while loop, Conditional statements are restricted to within properties. I sometimes feel we need papers like Dijkstra's "goto considered harmful" for visualization programming, I know how a visualization should look like but I still don't know how a program to generate that visualization should look like. We have tried to fit visualization programming into traditional programming paradigms (mostly oop) but perhaps we need a new machinery and a new programming paradigm to deliver flexible visualization tools quickly and efficiently.

Manas Mittal - Mar 09, 2011 03:26:25 am

Actually, I had a lot of questions about the Multi Scale Banking Paper.

  • How do we get the physical display slopes to be equal to s1/alpha and s2/alpha. I tried to do a ratio rule but got s1*alpha and s2*alpha.
  • Page2, last para, we can apply equation (3) locally instead of globally. What effect does that have? Do we lose anything?
  • In section 3.1, why did the authors consider multiply by hamming distance for windowing?

Karl He - Mar 09, 2011 02:08:52 pm

The discussion about graphs being opinionated interested me the most. I had always thought not showing 0 or using breaks was a horrible idea, since it distorts how you perceive the data, but since visualization is always opinionated doing those things becomes more of an issue of finesse. What matters the most is showing what you intended to show.

I think using space effectively basically follows from this. If you are focusing on what you are trying to show in your visual, the better a job you do the more the space is being used effectively.



[add comment]
Personal tools